HI! Would anybody suggest what function/package/method can I use to obtain forecasts for horizons longer than 1 period ahead? I want to do multistep forecasts with multilayer feedforward networks. I have tried several packages, but it turned out that the functions yield only one-step forecasts. To my best knowledge, *neuralne*t and *AMORE* packages don;t allow forecasting (you need to supply new input data for the forecast period and a function sim() from AMORE package gives you 1-step aheads). *nnet* has only 1 hidden layer, while in *RSNNS* I have tried predict.rsnns(), but having used mlp for training I get only binary result (e.g. a vector of "1"s). Regards, [[alternative HTML version deleted]]